Abstract
In this paper, we develop a first-order and a second-order coupled energy stable numerical scheme respectively for a Q-tensor based hydrodynamic model of nematic liquid crystal flows. We then extend the first order coupled scheme to a decoupled scheme and show that it is energy stable as well. The fully coupled schemes are implemented in 2-dimensional space and time, with which we study defect dynamics in flows of nematic liquid crystals in a channel. The numerical schemes are shown to be efficient in solving the Q-tensor based liquid crystal model. The methodology developed here also provides a paradigm for developing energy-stable schemes for more general hydrodynamic models of complex fluids which obey an energy dissipation law.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Background
Nematic liquid crystals are complex fluids that exhibit an orientational order on average in their molecular orientation, but do not normally possess any positional order. Like other complex fluids, their rheological properties are primarily dominated by mesoscopic structures and dynamics. The most popular mathematical model for flows of low molecular weight nematic liquid crystals is the celebrated Ericksen–Leslie model [13], in which the average molecular orientation is described by a unit vector \(\mathbf p\), known as the director, and the distortional elasticity is described by the Oseen–Frank elastic energy. This theory is apolar in that it does not differentiate the orientation between \(\mathbf p\) and \(-\mathbf p\). It describes uniaxial symmetry in all flow geometries including the shear flow and at defects, neglecting any potential biaxiality. However, in shear flows, orientational symmetry can readily be broken by the flow, leading to truly biaxial symmetry in nematic liquid crystals. In this case, the Ericksen–Leslie theory fails to capture this important broken symmetry. In fact, it has been show that the director model is in fact an asymptotic limit of a more general tensor model for liquid crystals [28].
An alternative to the director theory is to use the Q-tensor, a second order tensor of trace zero, which can be traced back to the deviatoric part of the second moment of a probability distribution function for the nematic liquid crystal system. The reflective symmetry of the system as well as biaxiality are naturally built in the tensor-based theories. Therefore, for flows of nematic liquid crystal polymers, a Q-tensor based hydrodynamic model of liquid crystals is more appropriate [1, 8, 9, 25, 28].
In the Q-tensor based Landau–de Gennes theory, the average orientation of nematic liquid crystals is described using \(\mathbf {Q}\) [28]. We denote
A nematic liquid crystal is said to be (i) isotropic when \(\mathbf {Q}=0\); (ii) uniaxial when \(\mathbf {Q}\) has two equal non-zero eigenvalues:
and biaxial when \(\mathbf {Q}\) has three distinct eigenvalues:
Here the order parameters: \(s_1, s_2\) are actually confined in a triangular region enclosing the origin [27]. They are linear combinations of eigenvalues of \(\mathbf Q\). In nontrivial flows of nematic liquid crystals, all three phases may show up.
Now, we consider nematic liquid crystals in a smooth domain \(\Omega \subset \mathbb {R}^3\). The total free energy F is given as follows:
with K the distortional elastic constant and the second term \(g(\mathbf {Q})\) is the bulk free energy term. Although more general distortional elastic terms can be included, for the sake of simplicity, we stick to the one constant case in this paper [14]. We use the simplified Landau–de Gennes bulk free energy [4]
where \(a_{1,2,3}\) are model parameters for the bulk energy. For thermotropic liquid crystals, \(a_1 \sim (T-T_0)\), \(T_0\) is the phase transition temperature, and T the temperature of the liquid crystal, which controls the phase transition from the isotropic phase to the nematic phase. I.e. when \(T_0 < T\), the equilibrium state is isotropic; when \(T_0 > T\), it is nematic. For lyotropic liquid crystals, \(a_1=\alpha (1-N/3), a_2=-\alpha N, a_3=\alpha N\), where N is the dimensionless concentration and \(\alpha \) is a scaling parameter [6, 28]. When, \(0\le N<3\), the system is in isotropic phase while it is in nematic phase when \(N>3\). There could also be an intermediate interval in N where both phases can coexist [27].
We denote the rate of strain tensor \(\mathbf {D}\) and the vorticity tensor \(\mathbf {W}\) as follows
with \((\nabla {\mathbf {u}})_{ij} = \partial _j {\mathbf {u}}_i\). The nondimensionalizsed Q-tensor model of nematic liquid crystals is summarized as follows [1, 28, 34]
where \(p'\) is the hydrostatic pressure, \(\sigma '\) is the elastic stress, \(\mathbf {H} \) is the molecular field, \(1/M_1\) is the relaxation time, \(S(\nabla {\mathbf {u}}, \mathbf {Q})\) is defined by
a is a geometric parameter of the nematic liquid crystal molecule, confined between -1 and 1 [28]. Here the molecular field \(\mathbf {H} \) is given by
and the elastic stress \(\sigma '\) is given by
This model has been used in studying liquid crystals especially flows of liquid crystal polymers extensively [4, 5, 16, 22, 28, 32]. Despite that the model has been solved numerically repeatedly, no numerical analysis result has been established. There are some existing analytic work in literature on solution existence and regularity. In [10], the author showed the existence and uniqueness of a local-in-time weak solution with weak regularity for the time derivative of the velocity and the tensor variable (\({\mathbf {u}}, \mathbf {Q}\)). In [7], the authors showed regularity criteria for local strong solutions. In [17], the authors proved the existence of global weak solutions in dimensions two and three in the entire space. In [11], the author considered the existence and uniqueness of weak solutions in a bounded three-dimensional domain under a homogeneous Dirichlet boundary condition for \({\mathbf {u}}\) and either a non-homogeneous Dirichlet or a homogeneous Neumann boundary condition for tensor \(\mathbf {Q}\). Numerically, a few authors have designed numerical schemes to solve the equations. For instance, in [5], Yeomann et al. used the Lattice Boltzmann method to solve the model to study structures in nematic liquid crystals; in [15], the author presented numerical schemes for minimizing Landau–de Gennes energy based on a finite element discretization without hydrodynamics. There are several extensions of this model to study two phase fluid flows [23] and active liquid crystals recently [2]. However, no stability analysis has been established for the numerical schemes developed so far.
In fact, it can be shown that the Q-tensor nematic liquid crystal model (1.7) obeys an energy dissipation law, which will be presented in details in the next section. This indicates that the model can be derived using the generalized Onsager principle [33]. Naturally, one would like to design numerical schemes that approximate the continuum model and in the meantime obey consistent discrete energy dissipation laws, which would be a good indication of a fine approximation to the continuum dissipative system. This type of numerical schemes is known as the energy stable scheme. However, to our best knowledge, such a numerical scheme for the Q-tensor hydrodynamic model of liquid crystals (1.7) has not been developed so far.
In this paper, we will develop several energy stable numerical schemes for this nematic liquid crystal model (1.7), based on a stabilization technique [20, 30], the convex splitting strategy [19, 26] and decoupling methods [3, 21]. We will only discretize the scheme in time. For the spatial discretization, we can readily develop a second order finite difference discretization to retain the discrete energy in discrete norms under appropriate boundary conditions. For finite element and spectral methods, energy preserving discretization perhaps can be established as well, which will not be pursued in this paper. Thus, the numerical schemes can be validated for a broad spectrum of spatial-discretization strategies so long as the spatial discretization renders a faithful approximation to the total energy in the continuous system. The numerical schemes that we will develop in this study will address the following numerical issues directly:
-
the coupling of the velocity and pressure through the incompressible condition;
-
the stiffness in the tensor equation associated with nonlinear bulk terms \(g(\mathbf {Q})\) (an explicit treatment of \(g(\mathbf {Q})\) will lead to instability);
-
the nonlinear coupling between the momentum transport and the tensor field equation.
The rest of the paper is organized as follows. In Sect. 2, we briefly derive the energy dissipation law of the nematic liquid crystal model (1.7). In Sect. 3, we present several semi-discrete numerical schemes for the model and prove their energy stability. Fully discretized schemes of course warrant thorough numerical analyses to prove their stability and convergence, which will be deferred to a future study. In Sect. 4, we present several numerical case studies on mesh-refinement and defect dynamics of nematic liquid crystal channel flows. We then give a concluding remark in the last section.
2 Energy Dissipation in the Nematic Liquid Crystal System
In this section, we show that model (1.7) satisfies an energy dissipation law. In order to simplify our proof, we rewrite the hydrostatic pressure p and elastic stress \(\sigma \) as
Then, the model is rewritten into
with
In the following discussion, we will focus on deriving energy-stable numerical schemes on (2.3).
We consider a smooth domain \(\Omega \). For any two vector function \(\mathbf {f}=(f_1,f_2,\ldots , f_d)\) and \(\mathbf {g}=(g_1,g_2,\ldots , g_d)\), where \(d \in N\), the inner product and norm are defined by
First, we present the following lemmas.
Lemma 2.1
The following identity holds for any vector function \({\mathbf {u}}\),
provided that \(\mathbf{n} \cdot \mathbf {v}|_{\partial \Omega } = 0\), \(\mathbf {v}\) and \( {\mathbf {u}}\) are sufficiently smooth and \(\nabla \cdot \mathbf {v}= 0\), where \(\mathbf n\) is the unit external normal of the boundary \(\partial \Omega \).
Lemma 2.2
The following identity holds for any scalar function p,
provided that \(\mathbf{n} \cdot \mathbf {v}|_{\partial \Omega } = 0\), \(\mathbf {v}\) is sufficiently smooth and \(\nabla \cdot \mathbf {v}= 0\), where \(\mathbf n\) is the unit external normal of the boundary.
The proof of Lemma 2.1 and 2.2 are straight forward. Readers can refer to [24] for details.
Lemma 2.3
Given a smooth domain \(\Omega \), if \({\mathbf {u}}\in \mathbb {R}^3\) with \({\mathbf {u}}|_{\partial \Omega } =0 \), and \(\mathbf {Q}, \mathbf {H} \in \Lambda \), the following identity holds
where \(\sigma (\mathbf {Q}, \mathbf {H} )\) is defined in (2.5 2.6) and \(S(\nabla {\mathbf {u}}, \mathbf {Q})\) in (2.4).
Proof
Recall from (2.4) and (2.5 2.6) that
For the left hand side, we have
provided \({\mathbf {u}}|_{\partial \Omega } = 0\).
For the right hand side, we have
Notice the facts that
Adding the left hand side up, we obtain \((\nabla \cdot \sigma , {\mathbf {u}})\). Analogously, adding the right hand side, we have \((\mathbf {H} , S(\nabla {\mathbf {u}}, \mathbf {Q}))\).
So, given \({\mathbf {u}}|_{\partial \Omega } = 0\), we arrive at
\(\square \)
Lemma 2.4
Recall the Landau–de Gennes bulk potential in (1.5). We can rewrite it as
where
If \(b_1 > \max (|a_2|,|a_1|)\) and \(b_3 > |a_2|\), \(f_1(\mathbf {Q})\) and \(f_2(\mathbf {Q})\) are convex functions.
Proof
We only need to prove the Hessian matrix of \(f_1\) and \(f_2\) are positive-definite. For notation simplicity, we use the index notations, i.e. \((\mathbf {Q})_{ij}=Q_{ij}\), and we use the notational convention of Einstein summation.
For this bulk potential
we can get
And the Hessian matrix could be calculated as
If we multiply left side of \(\mathcal {H}_1\) by \(x_{ij}\), right side of \(\mathcal {H}_1\) by \(x_{mn}\), we obtain,
since \(b_1 > max(|a_2|,|a_1|)\) and \(b_3 > |a_2|\). Therefore, \(f_1(\mathbf {Q})\) is a convex function. In the same manner, it is easy to show \(f_2(\mathbf {Q})\) is also a convex function. \(\square \)
The energy dissipation law of model (2.3) is summarized in the following theorem.
Theorem 2.1
We denote the total energy as \(\mathscr {E}\)
where the first term \(\frac{1}{2}{\mathbf {u}}^2\) is the kinetic energy density and f is the free energy density given by
The energy dissipation rate (or negative of the entropy production rate) of system (2.3) is given by
Proof
The time rate of change for the total energy is calculated as follows:
Notice from (2.3) that
By substituting (2.26) into (2.25), we obtain
Using Lemmas 2.1, 2.2 and 2.3, we have respectively
and
Finally, we end up with
\(\square \)
This energy dissipation property provides the foundation, as well as a guidance for us to derive semi discrete energy-stable schemes for the nematic liquid crystal model. Specifically, in the next section, we will propose a first-order and a second-order unconditional, energy-stable scheme for this model, respectively. Then, we will discuss decoupling strategies to yield two linearly, first order, decoupled energy stable schemes for the same model. The proof of stability of the schemes will be given in details in the following.
3 Semi-Discrete Numerical Schemes
To design numerical schemes and prove their energy stability, we have to assume some properties for the bulk potential function \(g(\mathbf {Q})\) technically, i.e. it satisfy the following properties: (i). g has a continuous second order derivative, (ii). there exists a constant L such that
where \(\mathcal {H}(\mathbf {Q})\) is the Hessian matrix of \(g(\mathbf {Q})\). One might immediately notice, the Landau–de Gennes bulk potential \(g(\mathbf {Q})\) in (1.5) does not satisfies (3.1) as it is.
In practice, the properties can be readily met provided we modify the given function of \(g(\mathbf {Q})\) far away from the origin by an essentially quadratic potential of \(\mathbf Q\) to meet the conditions. Physically, \(\mathbf {Q}\) is bounded since it is the deviatoric part of a second moment of a probability density function and the bulk potential makes sense only for \(\mathbf Q\) being confined in its physically relevant domain. Thus, any modification of the bulk potential beyond the domain would not change the physical relevance of the model. For instance, we can propose
In addition, the bulk-potential \(g(\mathbf {Q})\) could be splitted into a convex part \(g_c(\mathbf {Q})\), minus another convex part \(g_e(\mathbf {Q})\) [29], i.e.
One case is given in (2.16) and verified in Lemma 2.4. Notice to make our proof general, we don’t necessarily restrict to any specific bulk potentials in the following discussion.
We next introduce the following notations:
We use \(E^n\) to denote the discrete energy at time step \(t^{n}\),
and denote
where \(\mathcal {H}(\mathbf {Q})\) is the Hessian matrix of \(g(\mathbf {Q})\).
3.1 Preliminaries
We first present some lemmas to assist readers to navigate through the details of the proof of the main theorem.
Lemma 3.1
The following equalities hold,
Proof
We can obtain the equalities by simply expanding the inner product on the right hand side and then combining the corresponding terms. \(\square \)
Lemma 3.2
Let \(g \in C^2(\Lambda )\), with \(\Lambda \) defined by (1.1).
-
(i)
If
$$\begin{aligned} \max _{\mathbf {x}\in \Lambda } |\mathcal {H}(\mathbf {x})| < L, \end{aligned}$$(3.8)
where \(H(\mathbf {x})\) is the Hessian matrix of \(g(\mathbf {x})\), then \(\forall \mathbf{x}^{n+1} , \mathbf{x}^n \in \Lambda \), the following inequality holds,
-
(ii)
If g is a convex function, i.e. \(H(\mathbf {x})\) is positive definite \(\forall \mathbf {x}\in \Lambda \), the following two inequalities hold
$$\begin{aligned} - \left( \mathbf {x}^{n+1}-\mathbf {x}^n, \nabla _{\mathbf {x}} g (\mathbf {x}^n) \right)\ge & {} - \left( g(\mathbf {x}^{n+1}) - g(\mathbf {x}^n),1 \right) \end{aligned}$$(3.10)$$\begin{aligned} \left( \mathbf {x}^{n+1}-\mathbf {x}^n, \nabla _{\mathbf {x}} g(\mathbf {x}^{n+1}) \right)\ge & {} \left( g(\mathbf {x}^{n+1}) - g(\mathbf {x}^n),1 \right) \end{aligned}$$(3.11)
Proof
Using the Taylor expansion, \(\forall \mathbf {x}^n, \mathbf {x}^{n+1} \in \Lambda \), we have
-
(i)
If (3.8) holds, from (3.12), we have
$$\begin{aligned} g(\mathbf {x}^{n+1}) \le g(\mathbf {x}^n) + \nabla _{\mathbf {x}} g(\mathbf {x}^n) (\mathbf {x}^{n+1} - \mathbf {x}^n) + L |\mathbf {x}^{n+1} - \mathbf {x}^n|^2. \end{aligned}$$(3.14)
Integrating over \(\Omega \), we obtain
-
(ii)
If g is a convex function, from (3.12), we obtain
$$\begin{aligned} g(\mathbf {x}^{n+1})-g(\mathbf {x}^n)) \ge (\mathbf {x}^{n+1}-\mathbf {x}^n) \cdot \nabla _{\mathbf {x}} g(\mathbf {x}^n), \end{aligned}$$(3.16)
Taking intergation over \(\Omega \), we will arrive
Analogously, from (3.13), we can derive
\(\square \)
Lemma 3.3
Define
Then, the following inequality holds,
provided \(C_1 > L\), where \(E^{n+1}\) and \(E^n\) are the discrete free energy defined in (3.5).
Proof
If \(C_1 > L\), according to Lemma 3.2, we obtain
\(\square \)
Lemma 3.4
Define
where
Then,
Proof
We take the inner product of (3.22) with \(\mathbf {Q}^{n+1} - \mathbf {Q}^n\). Then, we obtain
\(\square \)
Lemma 3.5
Define
where
Here we denote \(g(\mathbf {Q}) = g_c(\mathbf {Q}) - g_e(\mathbf {Q})\) with \(g_c\) and \(g_e\) are all convex functions. Then,
Proof
Taking the inner product of (3.26) with \(\mathbf {Q}^{n+1} - \mathbf {Q}^n\), we have
since \(g_c\) and \(g_e\) are convex functions. Using Lemma 3.2, we arrive at
\(\square \)
3.2 First-Order, Coupled Unconditionally Energy Stable Schemes
Here, we present a first-order, coupled, linear, unconditionally energy stable scheme.
Given the initial condition \(\mathbf {Q}^0, {\mathbf {u}}^0\) and \(p^0=0\) and having computed \((\mathbf {Q}^n, {\mathbf {u}}^n, p^n)\), we calculate \((\mathbf {Q}^{n+1}, {\mathbf {u}}^{n+1}, p^{n+1})\) in the following two steps,
- (1) :
-
Step 1: update \((\mathbf {Q}^{n+1}, {\mathbf {u}}_\star ^{n+1})\):
$$\begin{aligned} \left\{ \begin{array}{l} \frac{\mathbf {Q}^{n+1} - \mathbf {Q}^n}{\delta t} + {\mathbf {u}}_\star ^{n+1} \cdot \nabla \mathbf {Q}^{n} -S \left( \nabla {\mathbf {u}}^{n+1}_\star , \mathbf {Q}^n\right) = M_1 \mathbf {H} ^{n+1},\\ \mathbf {H} ^{n+1} = - C_1 \left( \mathbf {Q}^{n+1} - \mathbf {Q}^n \right) + K\nabla ^2 \mathbf {Q}^{n+1} - g' \left( \mathbf {Q}^{n}\right) ,\\ \frac{{\mathbf {u}}_\star ^{n+1} - {\mathbf {u}}^n}{\delta t} + \left( {\mathbf {u}}^n \cdot \nabla \right) {\mathbf {u}}_\star ^{n+1} = \eta \Delta {\mathbf {u}}_\star ^{n+1} - \nabla p^n + \nabla \cdot \sigma \left( \mathbf {Q}^n,\mathbf {H} ^{n+1}\right) - \mathbf {H} ^{n+1} \nabla \mathbf {Q}^n \\ {\mathbf {u}}_\star ^{n+1}|_{\partial \Omega } = 0, \quad \frac{\partial \mathbf {Q}^{n+1}}{\partial \mathbf {n}}|_{\partial \Omega } = 0, \end{array} \right. \end{aligned}$$(3.31) - (2) :
-
Step 2: update \({\mathbf {u}}^{n+1}\):
$$\begin{aligned} \left\{ \begin{array}{l} \frac{{\mathbf {u}}^{n+1} - {\mathbf {u}}_\star ^{n+1}}{\delta t} = - \nabla \left( p^{n+1} - p^n\right) , \\ \nabla \cdot {\mathbf {u}}^{n+1} = 0, \quad {\mathbf {u}}^{n+1}\cdot \mathbf {n}|_{\partial \Omega } = 0. \end{array} \right. \end{aligned}$$(3.32)
Theorem 3.1
Given \(C_1 \ge L\), the scheme (3.31)–(3.32) obeys the discrete energy law
where \(E^n\) and \(E^{n+1}\) are defined in (3.5).
Proof
Taking the inner-product of (3.31) with \(2 \delta t {\mathbf {u}}_\star ^{n+1}\), we obtain,
To deal with the pressure term, we take the inner product of (3.32) with \(2\delta t^2\nabla p^n\) to arrive at
Taking the inner product of (3.32) with \(2 \delta t {\mathbf {u}}^{n+1}\), we obtain
It follows from (3.32) directly that
Combining (3.34)–(3.37), we obtain
If we take the inner product of (3.31) with \(2 \delta t \mathbf {H} ^{n+1}\), we get
Applying Lemma 2.3, we have
Then adding (3.38) and (3.39) and using the equality (3.40), we obtain
Applying Lemma 3.3, we finally arrive at
\(\square \)
Remark 3.1
In scheme (3.31)–(3.32), we use the first-order pressure correction scheme to handle the momentum transport equation together with the incompressibility condition. This scheme is thus first-order and energy stable. We notice that there exists a coupling between \(\mathbf {Q}^{n+1}\) and \({\mathbf {u}}_\star ^{n+1}\), which means that we have to solve these two equations simultaneously. We will design schemes to decouple them in the following.
Remark 3.2
\(C_1\) term is the added first-order stabilizer to balance the explicit treatment for \(g(\mathbf {Q})\) [21, 31]. One can also use the convex splitting strategy to deal with such a nonlinear term [19] without introducing the order one term.
3.3 Second-Order, Coupled Unconditionally Energy Stable Schemes
In a similar way, we now design a second-order, coupled, unconditionally energy stable scheme.
Given the initial condition \(\mathbf {Q}^0, {\mathbf {u}}^0\) and \(p^0=0\), we first compute \(\mathbf {Q}^1, {\mathbf {u}}^1\) and \(p^1\) by the first-order scheme (3.31)–(3.32). Having computed \((\mathbf {Q}^{n-1},{\mathbf {u}}^{n-1}, p^{n-1})\) and \((\mathbf {Q}^n, {\mathbf {u}}^n, p^n)\), where \(n \ge 2\), we calculate \((\mathbf {Q}^{n+1}, {\mathbf {u}}^{n+1}, p^{n+1})\) in the following two steps:
- (1) :
-
Step 1: update \((\mathbf {Q}^{n+1}, {\mathbf {u}}_\star ^{n+1})\):
$$\begin{aligned} \left\{ \begin{array}{l} \frac{\mathbf {Q}^{n+1} - \mathbf {Q}^n}{\delta t} + {\mathbf {u}}_\star ^{n+\frac{1}{2}} \cdot \nabla \overline{\mathbf {Q}}^{n+\frac{1}{2}} - S \left( \nabla {\mathbf {u}}_\star ^{n+\frac{1}{2}},\overline{\mathbf {Q}}^{n+\frac{1}{2}}\right) = M_1 \mathbf {H} ^{n+\frac{1}{2}} , \\ \mathbf {H} ^{n+\frac{1}{2}} = K \nabla ^2 \mathbf {Q}^{n+\frac{1}{2}} - {g'}^{n+ \frac{1}{2}} ,\\ \frac{{\mathbf {u}}_\star ^{n+1} - {\mathbf {u}}^n}{\delta t} + \left( \overline{{\mathbf {u}}}^{n+\frac{1}{2}} \cdot \nabla \right) {\mathbf {u}}_\star ^{n+\frac{1}{2}} = \eta \Delta {\mathbf {u}}_\star ^{n+\frac{1}{2}} - \nabla p^n + \nabla \cdot \sigma \left( \overline{\mathbf {Q}}^{n+\frac{1}{2}}, \mathbf {H} ^{n+\frac{1}{2}}\right) \\ \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad - \mathbf {H} ^{n+\frac{1}{2}} \nabla \overline{\mathbf {Q}}^{n+\frac{1}{2}}, \\ {\mathbf {u}}_\star ^{n+1}|_{\partial \Omega } = 0, \quad \frac{\partial \mathbf {Q}^{n+1}}{\partial \mathbf {n}}|_{\partial \Omega } = 0,{\mathbf {u}}_{\star }^{n+\frac{1}{2}} = \frac{1}{2} \left( {\mathbf {u}}_\star ^{n+1} + {\mathbf {u}}^n\right) , \end{array} \right. \end{aligned}$$(3.43)where \(\tilde{g}^{n+\frac{1}{2}}\) is a second-order approximation for \(g(\mathbf {Q}^{n+\frac{1}{2}})\). We will give a detailed discussion on choices of \(\tilde{g}\).
- (2) :
-
Step 2: update \({\mathbf {u}}^{n+1}\):
$$\begin{aligned} \left\{ \begin{array}{l} \frac{{\mathbf {u}}^{n+1} - {\mathbf {u}}_\star ^{n+1}}{\delta t} = - \frac{1}{2}\nabla \left( p^{n+1} - p^n\right) , \\ \nabla \cdot {\mathbf {u}}^{n+1} = 0, \quad {\mathbf {u}}^{n+1}\cdot \mathbf {n}|_{\partial \Omega } = 0. \end{array} \right. \end{aligned}$$(3.44)
Theorem 3.2
Given
the scheme (3.43)–(3.44) obeys the discrete energy law
where \(E^n\) and \(E^{n+1}\) are defined in (3.5).
Proof
The proof is similar to the one in the previous theorem. Nevertheless, we provide some details below. Taking the inner-product of (3.43) with \(2 \delta t {\mathbf {u}}_\star ^{n+\frac{1}{2}}\), we obtain,
To deal with the pressure term, we take the inner product of (3.44) with \(2\delta t^2\nabla p^n\) to arrive at
Taking the inner product of (3.44) with \(2 \delta t {\mathbf {u}}^{n+1}\), we obtain
It follows from (3.44) directly that
Noticing the fact
(due to \((\nabla p, {\mathbf {u}}^n)=0\)), and combining (3.47)–(3.50), we obtain
If we take the inner product of (3.43) with \(2 \delta t \mathbf {H} ^{n+\frac{1}{2}}\), we get
Applying Lemma 3.3, i.e.
and adding (3.52) and (3.53), we obtain
From Lemma 3.4, we have
Finally, we arrive at
\(\square \)
We can take advantage of the convex splitting strategy [12] to reduce the non-linear terms. Furthermore, we can also truncate the bulk potential and rewrite \(g(\mathbf {Q})\) as a convex function minus another convex function. This leads to the following theorem.
Theorem 3.3
If we define
the scheme (3.43)–(3.44) obeys the discrete energy law
with \(E^n\) and \(E^{n+1}\) given in (3.5).
Proof
This proof is similar to the one in Theorem 3.2 except that we need to use Lemma 3.5 instead of Lemma 3.4. We thus omit the details. \(\square \)
3.4 Linearly, First-Order Decoupled Energy Stable Schemes
Next, we discuss two ways to design decoupled energy stable schemes and remark on the difficulties to arrive at sharp results.
Given the initial condition \(\mathbf {Q}^0, {\mathbf {u}}^0\) and \(p^0=0\) and having computed \((\mathbf {Q}^n, {\mathbf {u}}^n, p^n)\), where \(n\ge 1\), we calculate \((\mathbf {Q}^{n+1}, {\mathbf {u}}^{n+1}, p^{n+1})\) in the following three steps,
- (1) :
-
Step 1: update \(\mathbf {Q}^{n+1}\):
$$\begin{aligned} \left\{ \begin{array}{l} \frac{\mathbf {Q}^{n+1} - \mathbf {Q}^n}{\delta t} + {\mathbf {u}}^n \cdot \nabla \mathbf {Q}^{n} - S\left( \nabla {\mathbf {u}}^n, \mathbf {Q}^n\right) = M_1 \mathbf {H} ^{n+1}\\ \mathbf {H} ^{n+1} = - C_1 \left( \mathbf {Q}^{n+1} - \mathbf {Q}^n \right) + K \nabla ^2 \mathbf {Q}^{n+1} - g' \left( \mathbf {Q}^{n}\right) ,\\ \frac{\partial \mathbf {Q}^{n+1}}{\partial \mathbf {n}}|_{\partial \Omega } = 0, \end{array} \right. \end{aligned}$$(3.60) - (2) :
-
Step 2: update intermediate variable \({\mathbf {u}}^n_\star \) and \(\Gamma ^n\):
$$\begin{aligned} \beta ^n= & {} \Vert - \mathbf {H} ^{n+1} \nabla \mathbf {Q}^n + \nabla \cdot \sigma \left( \mathbf {Q}^n, \mathbf {H} ^{n+1}\right) \Vert , \nonumber \\ {\mathbf {u}}_\star ^n= & {} {\mathbf {u}}^n - \delta t \mathbf {H} ^{n+1} \nabla \mathbf {Q}^n + \delta t \nabla \cdot \sigma \left( \mathbf {Q}^n, \mathbf {H} ^{n+1}\right) , \nonumber \\ \Gamma ^n= & {} - \frac{\delta t (\beta ^n)^2 }{\Vert {\mathbf {u}}_\star ^n \Vert ^2} {\mathbf {u}}_\star ^n. \end{aligned}$$(3.61) - (3) :
-
Step 3: update \({\mathbf {u}}^{n+1}\):
$$\begin{aligned}&\left\{ \begin{array}{l} \frac{\tilde{{\mathbf {u}}}^{n+1} - {\mathbf {u}}^n}{\delta t} + ({\mathbf {u}}^n \cdot \nabla ) \tilde{{\mathbf {u}}}^{n+1} = \eta \Delta \tilde{{\mathbf {u}}}^{n+1} - \nabla p^n + \nabla \cdot \sigma \left( \mathbf {Q}^n, \mathbf {H} ^{n+1}\right) - \mathbf {H} ^{n+1} \nabla \mathbf {Q}^n + \Gamma ^n\\ \tilde{{\mathbf {u}}}^{n+1}|_{\partial \Omega } = 0. \end{array} \right. \end{aligned}$$(3.62)$$\begin{aligned}&\left\{ \begin{array}{l} \frac{{\mathbf {u}}^{n+1} - \tilde{{\mathbf {u}}}^{n+1}}{\delta t} = - \nabla \left( p^{n+1} - p^n\right) , \\ \nabla \cdot {\mathbf {u}}^{n+1} = 0, \quad {\mathbf {u}}^{n+1}\cdot \mathbf {n}|_{\partial \Omega } = 0. \end{array} \right. \end{aligned}$$(3.63)
Theorem 3.4
Given \(C_1 > L\), the scheme given by (3.60)–(3.63) satisfies the following semi-discrete energy dissipation law:
where the semi-discrete energy \(E^{n}\) is defined in (3.5) and
Proof
From the definition of \({\mathbf {u}}_{\star }^n\) in equation (3.61), we rewrite the momentum equation as follows
If we take the inner product of (3.60) with \(2 \delta t \mathbf {H} ^{n+1}\), we get
Taking the inner product of (3.61) with \(2{\mathbf {u}}^n\), we obtain
From,
we arrive at
i.e.
Notice the fact \(\delta t^2 (\beta ^n)^2 = \Vert {\mathbf {u}}_\star ^n - {\mathbf {u}}^n \Vert ^2\) and if we denote
we obtain
Taking the inner-product of (3.68) with \(2 \delta t \tilde{{\mathbf {u}}}^{n+1}\), we obtain,
To deal with the pressure term, we take the inner product of (3.63) with \(2\delta t^2\nabla p^n\) to arrive at
Taking the inner product of (3.63) with \({\mathbf {u}}^{n+1}\), we obtain
It follows from (3.63) directly that
Combining (3.75) and (3.76)–(3.79), we obtain
Adding (3.69) with (3.70) and noticing the fact that
we arrive at
Finally, adding up the Eqs. (3.80), (3.82) and dividing both side by 2, we have
According to Lemma 3.3,
we finally obtain
\(\square \)
Remark 3.3
We note that, in the momentum equation, a term \(\Gamma ^{n}\) is added to cancel other terms generated from \(u_\star ^n\). This leads to the result that such a scheme is only “quasi”-energy stable, i.e. the energy decays modulo a quantity of order \(O(\delta t^4)\).
4 Numerical Results
In this section, we first conduct the mesh refinement test in time to verify the convergence rate of the first order and the second order coupled scheme, respectively. Then, we will conduct several numerical simulations of liquid crystal flows in 2D channel geometries focusing on flow-induced defect dynamics. Here, we consider a lyotropic liquid crystal system with \(\eta = 1\), \(M_1 = 1\), \(K=0.001\), \(\alpha =1, N=4\), and \(a = 0.8\), respectively, which represents rod-like nematic liquid crystals. We use a rectangular computational domain, denoted by \([0, Lx] \times [0, Ly]\) in all the simulations below. In the following numerical study, we set \(Lx=Ly=1\), and we use periodic boundary condition in the x direction. In the y-direction, the boundary conditions are given as follows:
The space is discretized by a second order finite difference method with central differencing. We use staggered grids, where \(\mathbf {Q}\) and p are discretized at the cell center and \({\mathbf {u}}\) is discretized on cell edges. A schematic cartoon is shown in Fig. 1 to illustrate this idea.
4.1 Time Step Refinement Test
First, we conduct a mesh refinement test in time to confirm the order of the schemes. Notice that the schemes are coupled between the velocity field \(\mathbf {v}\) and nematic field \(\mathbf {Q}\). In the numerical implementation, we employ the extrapolation technique to decouple them. This should not affect any temporal accuracy. In order to have a high resolution in space to minimize the influence of the spatial error, we test the code in 2D spatial domain with the spatial mesh-size: \(512 \times 512\) and time step \(\delta t = 2\times 10^{-3}\), \(10^{-3}\), \(5 \times 10^{-4}\), \(2.5 \times 10^{-4}\) and \(1.25 \times 10^{-4}\), respectively. At \(t=1\), the numerical solutions are compared, where we calculate the error by treating the numerical result with its nearest finer time step as the approximation to the accurate solution. The errors for the first-order scheme in \(L_1, L_2\) and \(L_{\infty }\) norm for \(\mathbf {v}\) and \(\mathbf {Q}\) are shown in Fig. 2, respectively. From the numerical tests, the first order scheme is shown at least first-order accurate in time. The second-order scheme is also implemented and the mesh refinement test result is depicted in Fig. 3. From the numerical test, the numerical scheme is shown as second-order accurate.
We next apply this numerical scheme to study defect dynamics in liquid crystal flows during coarsening and defect annihilation.
4.2 Coarsening of Liquid Crystal Structures
In this simulation, we use the initial condition of a nematic liquid crystal phase on the xy plane, shown in Fig. 4a, where several defects are present. The coarsening dynamics of the orientational structure involving multiple defects are simulated. Several plots at various time steps are shown in Fig. 4a–g. We also plotted the discrete energy as a function of time in Fig. 4h, where we observe the energy decay with respect to time. Although there does not exist active hydrodynamic impact to the flow, its effect is coupled to liquid crystal orientational dynamics. This figure shows very interesting coarsening dynamics, where we observe \(+1\) defects divide into two \(+\frac{1}{2}\) defects (comparing Fig. 4a, b), as well as the annihilation of \(-\frac{1}{2}\) and \(+\frac{1}{2}\) defects, which can also be seen from the eigenvalues of \(\mathbf {Q}\) on the xy plane, plotted in Fig. 5a–c. The velocity field and vorticity field are also shown in Fig. 5, from which, we observe relatively larger vorticity near the boundary of defects, as coarsening dynamics evolves. Initially there exists 10 defects, 4 degree \(\pm 1\) and 6 degree \(\pm 1/2\). Eventually, the defects disappear due to annihilation. During the process, very complex defect dynamics are captured by the simulation.
4.3 Annihilation and Repulsions of a Single Defect Pair
To better understand coarsening dynamics involving point defects, we conducted two more simulations, where annihilation of a \(+\frac{1}{2}\) and \(-\frac{1}{2}\) defects and the repulsion of two \(+\frac{1}{2}\) defects are shown in Figs. 6 and 8, respectively. In both cases, we see decay of discrete energies with time (see Figs. 6f and 8f), which demonstrates the robustness of our energy-stable schemes. In particular, we have plotted the time series of the difference between the largest eigenvalue and the second largest one of \(\mathbf {Q}+\mathbf{I}/3\) in Fig. 7a–c and the difference between the second largest eigenvalue and the smallest one in Fig. 7d–f, respectively. It demonstrates that the liquid crystal at the point defects are in fact biaxial [18] (as we have three distinct eigenvalues of \(\mathbf {Q}+ \mathbf {I}/3\) in the neighborhood of the point defects) and the liquid crystal is uniaxial elsewhere (as the second largest eigenvalue and the smallest eigenvalue are equal).
5 Conclusion
In this paper, we have developed several semi-discrete energy stable schemes for the Q-tensor based hydrodynamic model of nematic liquid crystals, which can be used to study flowing behavior of nematic liquid crystals in various geometries. These include both a first order and a second order scheme, together with a first-order decoupled approximation. The new schemes lay a solid foundation for pursuing further development of energy stable, efficient numerical methods for solving hydrodynamic models of complex fluids that obey energy dissipation laws. The schemes are then used to study defect dynamics during coarsening of nematic liquid crystal orientational structures and pair interaction between point defects. The numerical schemes are shown to be effective in solving the Q-tensor based hydrodynamic model of liquid crystals. The methodology developed here also provides a paradigm for developing energy-stable schemes for more general hydrodynamic models of complex fluids, which obey an energy dissipation law.
References
Beris, A.N., Edwards, B.: Thermodynamics of Flowing Systems. Oxford Science Publications, New York (1994)
Blow, M.L., Thampi, S.P., Yeomans, J.M.: Biphasic lyotropic active nematics. Phys. Rev. Lett. 113, 248303 (2014)
Boyer, F., Lapuerta, C.: Study of a three component Cahn–Hilliard flow model. ESAIM Math. Model. Numer. Anal. 40(4), 653–687 (2006)
de Gennes, P.G., Prost, J.: The Physics of Liquid Crystals. Oxford University Press, Oxford (1993)
Denniston, C., Orlandini, E., Yeomans, J.M.: Lattice boltzmann simulations of liquid crystal hydrodynamics. Phys. Rev. E 63(5), 056702 (2001)
Doi, M., Edwards, S.F.: The Theory of Polymer Dynamics. Oxford Science Publication, New York (1986)
Fan, J., Ozawa, T.: Regularity criteria for a coupled Navier–Stokes and q-tensor system. Int. J. Anal. 2013, 718173 (2013)
Forest, M.G., Wang, Q.: Hydrodynamic theories for mixtures of polymers and rodlike liquid crystalline polymers. Phys. Rev. E 72, 041805 (2005)
Forest, M.G., Wang, Q., Zhou, H.: Homogeneous biaxial patterns and director instabilities of liquid crystal polymers in axial and planar elongation. Phys. Fluids 12, 490–498 (2000)
Guillenn-Gonzalez, F., Rodriguez-Bellido, M.A.: Weak time regularity and uniqueness for a q-tensor model. SIAM J. Math. Anal. 46(5), 3540–3567 (2014)
Guillenn-Gonzalez, F., Rodriguez-Bellido, M.A.: Weak solutions for an initial boundary q tensor problem related to liquid crystals. Nonlinear Anal. 112, 84–104 (2015)
Han, D., Wang, X.: A second order in time uniquely solvable unconditionally stable numerical schemes for Cahn–Hilliard–Navier–Stokes equation. J. Comput. Phys. 290(1), 139–156 (2015)
Leslie, F.M.: The theory of flow phenomena in liquid crystals. Adv. Liq. Cryst. 4, 1–81 (1979)
Longa, L., Trebin, H.R.: Spontaneous polarization in chiral biaxial liquid crystals. Phys. Rev. A 42(6), 3453 (1990)
MacDonald, C.S., Mackenzie, J.A., Ramage, A.: Efficient moving mesh method for q-tensor models of nematic liquid crystals. SIAM J. Sci. Comput. 37(2), 215–238 (2015)
Marenduzzo, D., Orlandini, E., Yeomans, J.M.: Hydrodynamics and rheology of active liquid crystals a numerical investigation. Phys. Rev. Lett. 98, 118102 (2007)
Paicu, M., Zarnescu, A.: Energy dissipaiton and regularity for a coupled Navier–Stokes and q-tensor system. Arch. Ration. Mech. Anal. 203, 45–67 (2012)
Schopohl, N., Sluckin, T.J.: Defect core structure in nematic liquid crystals. Phys. Rev. Lett. 59(22), 2582 (1987)
Shen, J., Wang, C., Wang, S., Wang, X.: Second-order convex splitting schemes for gradient flows with Ehrlich–Schwoebel type energy: application to thin film epitaxy. SIAM J. Numer. Anal. 50(1), 105–125 (2012)
Shen, J., Yang, X.: Numerical approximation of Allen–Cahn and Cahn–Hilliard equations. Discrete Contin. Dyn. Syst. Ser. B 28(4), 1669–1691 (2010)
Shen, J., Yang, X.: Decoupled energy stable schems for phase field models of two phase complex fluids. SIAM J. Sci. Comput. 36(1), 122–145 (2014)
Sonnet, A.M., Maffettone, P.L., Virga, E.G.: Continuum theory for nematic liquid crystals with tensorial order. J. Non-Newton. Fluid Mech. 119, 51–59 (2004)
Sulaiman, N., Marenduzzo, D., Yeomans, J.M.: Lattice boltzmann algorithm to simulate isotropic–nematic emulsions. Phys. Rev. E 74, 041708 (2006)
Temam, R.: Navier–Stokes Equations: Theory and Numerical Analysis. American Mathematical Society, Providence (2001)
Tsuji, T., Rey, A.D.: Effect of long range order on sheared liquid crystalline materials, part I: compatability between tumbiling and behavior and fixed anchoring. J. Non-Newton. Fluid Mech. 73, 127–152 (1997)
Wang, C., Wise, S.M.: An energy stable and convergent finite-difference scheme for the modified phase field crystal equation. SIAM J. Numer. Anal. 49, 945–969 (2011)
Wang, Q.: Biaxial steady states and their stability in shear flows of liquid crystal polymers. J. Rheol. 41, 943–970 (1997)
Wang, Q.: A hydrodynamic theory of nematic liquid crystalline polymers of different configurations. J. Chem. Phys. 116, 9120–9136 (2002)
Wise, S.M., Wang, C., Lowengrub, J.S.: An energy-stable and convergent finite-difference scheme for the phase field crystal equation. SIAM J. Numer. Anal. 47(3), 2269–2288 (2009)
Yang, X.: Error analysis of stabilized semi-implicit method of Allen–Cahn equation. Discrete Contin. Dyn. Syst. Ser. B 11, 1057–1070 (2009)
Yang, X., Feng, J.J., Liu, C., Shen, J.: Numerical simulations of jet pinching-off and drop formation using an energetic variational phase-field method. J. Comput. Phys. 218, 417–428 (2006)
Yang, X., Forest, M.G., Mullins, W., Wang, Q.: 2-D lid-driven cavity flow of nematic polymers: an unsteady sea of defects. Soft Matter 6, 1138–1156 (2010)
Yang, X., Forest, M.G., Wang, Q.: Near equilibrium dynamics and one-dimensional spatial–temporal structures of polar active liquid crystals. Chin. Phys. B 23(11), 118701 (2014)
Yang, X., Wang, Q.: Capillary instability of axisymmetric active liquid crystal jets. Soft Matter 10(35), 6758–6776 (2014)
Acknowledgments
Jia Zhao and Qi Wang are partially supported by NSF-DMS-1200487 and NSF-DMS-1517347, NIH-2R01GM078994-05A1, AFOSR-FA9550-12-1-0178, and an SC EPSCOR/IDEA award. In addition, Jia Zhao is also supported by a Dissertation Fellowship from the Provost Office of USC.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Zhao, J., Wang, Q. Semi-Discrete Energy-Stable Schemes for a Tensor-Based Hydrodynamic Model of Nematic Liquid Crystal Flows. J Sci Comput 68, 1241–1266 (2016). https://doi.org/10.1007/s10915-016-0177-x
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10915-016-0177-x